benchmark.asm

    TIA 0
    TIY 0
zeroLoop:    
    AM
    AIY 1
    CIY F
    JUMP zeroLoop    
mainLoop:
    TIA 1
    TIY -1
    DEM+
    TIY -7 ; this will take a LONG time in the stock system
    MA
    CIA 1
    JUMP mainLoop
    AO
    ENDS
END: JUMP END    
; 100,000 iterations = 26 sec on fast mode; 1 sec in forkphorus; 0.1 sec in turbowarp
; 7 instructions in main loop
; fast mode: ~27,000 instructions / sec
; forkphorus fast: ~700,000 instructions / sec
; turbowarp fast: ~7,000,000 instructions / sec
; original ran at 400 kHz, 6 cycles per instruction, so at most 66,666 instructions / sec
; but the original ran a VM, so the trainer must have been an order of magnitude faster
; so, in fast mode on Scratch we get better performance than the original